Portable by design
Permitted working context lives above any single model provider or chat product, so the research effort transcends changes of model or interface.
AI infrastructure for collaborative mathematical research
Open Math gives individuals and research groups ownership of the context created around their work. Collaborate with people and coding agents without tying the research record to a single model, provider, or chat product.
Pilot toolchain
Open Math treats context as a governed research asset—not as exhaust trapped inside an AI product. Private work stays private; deliberate contributions build a common research body that can outlast the current model, interface, and team. The collective context contributes to future innovation.
Permitted working context lives above any single model provider or chat product, so the research effort transcends changes of model or interface.
Project owners define membership. Each member decides, by default, what to contribute from private work to the shared project context.
Reviewed source, conclusions, and shared project context remain in the project repository: attributable, inspectable, and exportable.
A planned central gateway will carry authorized project context to selected open-weight models as they advance.
Open Math brings human collaboration and agent-assisted work into one project context. Discuss a question with another researcher, or invite a bounded project agent to work with the relevant code, data, and shared knowledge.
Each member keeps a personal checkout and private working context. When something is useful, they can share it with the group or work with a Project Librarian, Code, Data, Quant, TeX, or Proof agent. The people and project retain ownership of the research record—not the tool or its underlying model.
Explore with a personal agent in your own checkout and context scope.
Work with other members and invite the right project agent, granting it bounded access to the relevant code, data, or discussion.
Members decide which conclusions and working context should join the project memory; code and artifacts remain separately reviewable.
The next collaborator or project agent can begin from what the group already knows.
Open-weight access · Gateway in development
Open Math decouples the LLM from the research effort. A central, OpenAI-compatible gateway is being built to provide transparent, project-scoped access to selected hosted—and later self-hosted—open-weight models. As the models advance, authorized context moves with the project; routing, budgets, and usage attribution remain visible to its owners.
Future-proof by design: context transfer is part of the interface, not a manual migration between models.